urnetwork/build v2026.9.14-1045806490

on GitHub

Changes in v2026.9.14-1045806490

v2026.9.13-1045655440 ... v2026.9.14-1045806490 -- 35 commits across 5 components.

build

urnetwork/build 7d4ee1c83...e0d1ecfcc

• 2026.9.13-1045655440 restore android pin (defbf35be)
• Require all attempted release components to succeed (fd102d0b7)
  Make iOS and macOS build, export, validation, and publication failures fatal. Reject missing or empty artifacts and require complete selected desktop package outputs. Preserve explicit noncomponent opt-ins. Refs urnetwork/connect#211.

sdk

urnetwork/sdk 5e59405d4...fdfcc60a9

• Raise iOS extension artifact ceiling to 48 MiB (fdfcc60a9)
  Increase only the compiled extension executable budget and add deterministic synthetic boundary, override, and FIPS-policy tests. Preserve runtime memory limits and the other archive ceilings.

connect

urnetwork/connect 25c48d68b...e51bfe96e

• Report: the window rule measured a resend interval, not the round trip (b4a1270d4)
• Report: measure against the local baseline, not a reporting host's configuration (3e72ef058)
• Report the throughput review, its corrections, and the window sizing result (434334457)
• Document open throughput program work (33176c0b4)
• PROVIDERFIXES.md: the two provider fixes, how to verify them, and the open ceiling (942f10021)
  Evidence, rig procedures with pass/fail criteria and false-fix matrices for eefa8d85 (upstream socket receive autotuning) and 04b44241 (release of sources whose TCP return stays unadmitted); the post-fix ~650 Mb/s ceiling investigation with everything ruled out; levers measured and rejected; a VPS verification of the flight-gate merge; measurement traps; and the remaining task list.
• ip: test that the upstream socket actually autotunes through the dial path (629d7816c)
  TestUpstreamTcpConnLeavesReceiveBufferToAutotuning only checks that configureUpstreamTcpConn leaves SO_RCVBUF alone. A fix that moves the explicit buffer instead of removing it -- for example into the dialer's Control, before connect -- passes it while still locking autotuning and pinning the window.
• ip: time the unreachable-source release with a test clock, not the wall (218f33ad8)
  The release tests proved "does not release" by waiting 250 ms and checking that nothing happened, which CODESTYLE.md rules out as a primary proof, and the positive tests depended on a real 50 ms abandon timeout elapsing.
• ip: put scaledPow2WindowSize's doc comment back on scaledPow2WindowSize (648fe8825)
  eefa8d85 inserted configureUpstreamTcpConn between scaledPow2WindowSize and its doc comment, so godoc attached the window-scaling description to the socket helper. Comment-only move; no code change.

sn

urfoundation/sn bb8eee21a...abae9a6a8

• Lock historical fleet receipt replay correction (abae9a6a8)
• Decode ethclient input in historical alias regression fixture (2a8cf0951)
• Design mainnet bootstrap and require deterministic regression coverage (d6d61e934)
• Format deterministic fleet history regressions (c50dbd956)
• Preserve historical fleet alias receipt formats during renewal replay (3c48657d2)
• List producer capture tests from the running binary (ff62e1cae)
• Validate capture ownership against growing source census (43cc2b218)
• Lock effective oracle restoration support for renewal (a59294e98)
• Accept effective original oracle restores during fleet renewal (536dba654)
• Lock strict owned-RPC runtime rendering (e31236859)
• Use owned-backed loopback proxy for staging discovery (cf4eae3d7)
• Rerender strict staging for owned RPC continuation (1469b14d5)

server

urnetwork/server a004cc9c4...5c9c554ae

• Bind evaluator smoke to frozen epoch policy (5c9c554ae)
• Fix evaluator score contract drift (a0e9db6c1)
• Complete monitor host-scope correctness (d0c8ad171)
  Preserve no-policy cancellation and probe error semantics, report unique exclusion counts, and withhold incomplete fleet recovery conclusions. Add deterministic regressions and document external-reference and HTTP-client boundaries. Refs urnetwork/connect#211; the provider CPU fix and direct regression are tracked in Connect f10a173 and ab74d62.
• Checkpoint latest monitor host-scope follow-up changes (651e386af)
• Checkpoint exact host observation exclusions and monitor coverage (a363ca43a)
• Extend migration monitor regression through version 674 (84e1299e3)
• Include public Alert classes in catalog coverage (72fff04fb)
  The deterministic AST regression reproduces omission of exported Class literals. Cover both internal finding.class and public Alert.Class fields without accepting computed or unrelated values.
• Checkpoint migration index contracts and catalog scanner coverage (8aae51927)
• Checkpoint bounded publisher and log observation reducers (8316bc311)
• Enforce single attempts in strict local test environments (d1746ce01)
• Reject noncanonical onboarding and sealed state encodings (47053c9ae)
• Checkpoint Mimir publisher and monitor telemetry (097dcc8d9)

Filtered out of this changelog: 7 merge commits (both sides are already listed), 1 version-stamp commit written by all/run.sh on every release, 1 ungoogle re-tag commit written by all/run.sh. Re-run all/changelog.py --from v2026.9.13-1045655440 --to v2026.9.14-1045806490 --filters= to include them.